【摘要】 为帮助运维人员提前发现未知风险,减少因异常风险带来的损失,提出多种特征融合的异常检测方法。对关键性能指标(KPI)进行多维度的特征提取,使用主成分分析方法 (PCA)进行降维,对降维后的数据按照时序模式,使用小波分解提取出高频特征与低频特征,使用极限梯度提升(XGBoost)模型进行异常检测。实验结果表明,该方法有较好的普适性、查全率和准确度较高,受试者工作特征(ROC)曲线也普遍优于其它模型。更多还原

【Abstract】 To help operation and maintenance personnel discover unknown risks in advance and reduce the loss caused by anomaly risks,a multi-feature fusion anomaly detection method was proposed.Multi-dimensional feature extraction was performed on the key performance indicators(KPIs),and the principal component analysis(PCA)method was used to reduce the dimensions.According to the time series pattern,the high-frequency features and low-frequency features were extracted by wavelet decomposition.The extreme gradient boosting(XGBoost)model was used to detect anomaly.Experimental results show that the proposed method has good universality,high recall and accuracy,and its receiver operating characteristic(ROC)curve is generally better than that of other models.更多还原